NPLAN vs GenAI Solutions: why AI without an engine cannot solve Supply Chain Planning

The market's new promise is simple: connect AI directly to the company's data and let planning happen. Ask about sales, get an answer. Ask about inventory, get a suggestion. Ask about next month, and AI returns a ready-made scenario.

This approach works well for queries. It works for productivity. It works for information access. But it does not solve supply chain planning.

LLMs are good at interpreting data. They can read tables, cross-reference information, identify patterns and generate natural language answers. For users accustomed to rigid dashboards and static reports, the experience is genuinely better.

You can ask how much was sold last week, request an explanation for a demand drop on a given SKU, or ask for a suggested action for a specific region. The answer arrives fast, contextualized and easy to consume.

This layer has real value. It improves the user experience, reduces friction in accessing information and helps democratize data consumption. The point is understanding what it is, and what it is not.

Supply chain planning is not a question with a single answer. It is a combinatorial problem that must simultaneously respect capacity constraints, lead times, multi-level bills of materials, demand-supply synchronization and period-to-period consistency.

An LLM can suggest a plan. It can write a plan. It can even justify a plan with convincing arguments. What it does not do is guarantee that the plan is viable. It does not recalculate material requirements, does not respect finite capacity, does not link orders across BOM levels, does not propagate a demand change down to raw material purchasing.

Robust planning requires two distinct layers with clear responsibilities. One does not replace the other.

There is also a more technical point that often goes unnoticed in the discussion. Processes like MRP, production planning and supply optimization are not queries. They are structured, repetitive calculations that are highly sensitive to consistency. They must be deterministic, auditable and reproducible: the same input should generate the same plan, today and three months from now.

This kind of logic belongs to a planning engine, not to a conversational interface. AI can make access easier, explain results, compare versions and suggest alternatives. But the calculation must continue to run on a system designed for it, with clear rules, explicit constraints and predictable behavior.

NPLAN starts from a native supply chain engine, with real finite capacity, multi-level material explosion and continuous integration between demand, supply and execution. Scenarios propagate automatically across layers, without parallel spreadsheets or manual reconciliation.

AI enters as an additional layer, not as a replacement. It explains the plan generated by the engine, helps the user understand variations, suggests adjustments and accelerates scenario exploration. The plan is born consistent. AI helps interpret, adjust and explore.

This model follows the principle of separating the decision engine from the interaction layer. The engine handles mathematical and operational consistency. AI handles clarity, context and experience. The two layers coexist but do not blur.

It is this separation that distinguishes a planning platform from a productivity tool with AI applied to supply chain data.

What has been discussed so far, from the limitations of pure AI to the need for an engine, is not merely conceptual. It is exactly how NPLAN was built.

The platform follows a three-layer architecture: data, supply chain engine and AI agents. Each layer has a specific role and none tries to do the other's job.

The engine is responsible for planning. It calculates, simulates and ensures consistency, considering finite capacity, multi-level material structures and the real constraints of the operation. It is what ensures the generated plan is viable, not merely plausible.

AI acts as the interface. The planner asks questions in natural language, explores scenarios and tests alternatives, while the engine ensures every answer is tied to the reality of the chain. The user interacts as if chatting, but what sustains the conversation is deterministic computation.

This eliminates the central weakness of AI-only approaches: the lack of consistency. The contrast becomes clear when comparing both architectures side by side.

In practice, the plan does not need to be corrected afterwards. It is born consistent.

In recent years, the idea of an autonomous supply chain has gained traction, in which planning, purchasing and inventory decisions would be made automatically by AI-based systems. The concept is appealing. It promises speed, efficiency and less operational effort.

There is, however, a clear limit. Supply chain decisions have direct financial impact and must be justifiable. Changing a production plan, anticipating a strategic purchase or redistributing inventory across plants are not just calculations. They are decisions that affect cost, service level and risk across the entire chain.

This raises a practical question: who is responsible for an automated decision when the outcome is not as expected? And, equally important, how do you reconstruct and explain the reasoning behind a decision made without human intervention? In practice, full autonomy tends to hit a simple, non-negotiable factor: accountability.

Nothing discussed so far means Gen-AI has no place in planning. Quite the opposite.

The problem is not using AI. It is using AI in the wrong place.

When properly applied, Gen-AI improves the planner's experience. It makes the process faster, more accessible and more exploratory. But it must operate on a structured foundation.

In practice, that means organizing data, calculations and decisions before exposing anything to a language model.

At NPLAN, this logic is applied directly. Planning data is pre-structured so that agents can answer complex questions without losing context.

In addition to data, the agents also have specific tools. They can perform actions such as recalculating scenarios or adjusting parameters, always invoking the right engine instead of trying to solve the problem on their own.

To ensure consistency, an orchestrator agent interprets the user's intent and routes each question to specialist agents such as inventory, capacity or replenishment. Each agent operates with its own context, data and tools, reducing ambiguity and avoiding generic answers.

Another important point is how the data is handled. The language model does not receive the full dataset. It operates with limited context and controlled samples. When required, the system accesses the real data in a structured, local way.

This makes the process more efficient and much safer.

How to ensure security when using Gen-AI in supply chain

Companies that take supply chain seriously cannot expose their operational, commercial and strategic data to public AI models. That is why using Gen-AI in a corporate environment must follow a different standard from casual use.

At NPLAN, this is treated as a central part of the architecture.

In practice, this materializes as follows:

Previous slideNext slide

Real example of agents, data and governance operating together in planning

In the end, Gen-AI does not replace planning. It makes planning more accessible, faster and more usable, provided it is connected to a system that guarantees consistency, control and security.

Before betting on GenAI as a path to planning, one single question is worth asking:

Is there an engine ensuring the plan actually works, or just an AI describing what should happen?